TODO list

Implementation in progress (in the GitHub version or in a branch):

• Power tests:
  • for 1 ROC curve with var() and cov() for the given ROC, instead of passing to power.roc.test.numeric.
  • for 2 unpaired ROC curves
• Confidence intervals for all coords return values (especially ppv, npv, and the threshold at given sp/se)

Small-scale project for the short- to medium-term

• Input: grouped data (table)
• Comparison of 1 ROC curve (against the diagonal)
• Implement 'mean' function to average ROC curves (Fawcett, section 8, algos 3 & 4)
• pAUC with DeLong
• ci.auc delong: with qgumbel to compute interval bounds rather than qnorm
• update.roc function to modify a parameter (response, predictor, formula, … ?) of a roc curve
• X plot (sensitivity and specificity as a function of the threshold)
• Matthews correlation coefficient in coords.
• Confidence interval of the best threshold (not of its SE/SP)

Large-scale project for the long term:

• Multiclass ROC curve:
  • CI for Multi-class AUC
  • Multiclass ROC (VUS)
• Multireader, multicase ROC analysis (see for instance this paper)
• Time dependent AUC (the survAUC CRAN package already provides a basic implementation)

If you would like a new feature to be added to the list, please open a new issue